Hugging Face's logo

Datasets:
junidude14
/
PBISC-data

Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
single-cell-rna-seq
bulk-rna-seq
diffusion
Libraries:
Datasets
pandas
Polars
Dataset card Data Studio Files
xet
 Community
1
junidude14 commited on
Commit
ba7af4c
·
verified ·
1 Parent(s): 5aa6fb6

Archive 4 checkpoints (1-4)

Browse files
Files changed (4) hide show
  1. paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/README.md +100 -0
  2. paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/logs/unblind_ages.log +30 -0
  3. paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/logs/v2_draws1000.log +124 -0
  4. paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/logs/v2_test_draws50.log +64 -0
paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/README.md ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Behçet bulk ↔ sc pair recovery (2026-10-04, CPU only)
2
+
3
+ **Question (user, 2026-10-04).** Can PBISC's paired LoRA be tried on Behçet's disease? This needs bulk–sc pairs.
4
+
5
+ **Data** (Zheng 2022 PNAS):
6
+ - bulk PBMC GSE198533: 9 BD + 10 HC, all male.
7
+ - sc PBMC GSE198616: 4 BD + 4 HC.
8
+ - The authors say 3 BD donors are in both, but not which. The datalake listed 0 resolved pairs
9
+ (`gblee/datalake/.../STUDY_READINESS.csv`, D01).
10
+
11
+ **Rule kept.** The datalake says identity must not be inferred from age or sex. So the pairs were found from
12
+ expression alone, and the published ages were compared only afterwards.
13
+
14
+ ## Result
15
+
16
+ | sc sample | blind expression match | rules that call it (of 9) | primary rule | published ages | tier |
17
+ |---|---|---|---|---|---|
18
+ | PBMC_BD_2 (GSM5953768) | **BDV4** (GSM5951805) | 8 | called, r 0.78 | 31 / 31 | expression call + same age |
19
+ | PBMC_BD_3 (GSM5953769) | **BDV5** (GSM5951806) | 6 | best and mutual best, r 0.56 < 0.66 | 58 / 58 | best + mutual in 9/9 rules + same age |
20
+ | PBMC_BD_4 (GSM5953770) | **BDV6** (GSM5951807) | 8 | called, r 0.75 | 41 / 41 | expression call + same age |
21
+
22
+ - **All three pairs are consistent across every rule.** Each of the three BD sc samples has the same best bulk under
23
+ all 9 statistic × gene-set rules, and that best bulk is always a mutual best.
24
+ - **The ages confirm it.** All three blind matches have exactly the sc donor's age. If bulks were picked at random,
25
+ the chance of that is 6 × 10⁻⁴ (`out/UNBLIND.json`).
26
+ - **The fourth BD sc sample has no match.** PBMC_BD_1 (age 30) has no bulk partner: its best r is 0.19–0.34 and it is
27
+ never mutual or called. This fits the authors' "3 of 4". The only 30-year-old BD bulk, BDU2, is the uveitis subtype,
28
+ and PBMC_BD_1 has vascular involvement.
29
+ - **The HC sc samples are not in the bulk.**
30
+ - PBMC_HC_1 and PBMC_HC_2 have no chrY expression, so they are female; the bulk donors are all male. GEO lists
31
+ PBMC_HC_2 as male, which the expression contradicts.
32
+ - PBMC_HC_3 (58) and PBMC_HC_4 (41) are male and share ages with bulk HC5 and HC7, but expression does not match
33
+ them (best r 0.25 and 0.42, never called).
34
+ - **Negative control.** Neither female sc sample is called by the primary rule or by any r_best or dc_margin rule.
35
+ The margin_z statistic calls PBMC_HC_1 → HC6 at every gene-set size, which is sex-discordant and therefore wrong.
36
+ margin_z was also the weakest statistic on the controls.
37
+ - **Genes behind the matches** (`out/match_gene_view.csv`: the matched bulk and sc are both at the same extreme of
38
+ their modality):
39
+ - PBMC_BD_2: S100B, FOLR3, ZNF683, HLA-DQB2.
40
+ - PBMC_BD_3: TMEM176A, PRSS21, MARCO.
41
+ - PBMC_BD_4: MDGA1, ZNF683, LGALS2, GLB1L.
42
+
43
+ **So:** Behçet has **3 bulk–sc pairs, all BD and all vascular type, and no HC pair**. The other 6 BD and 10 HC bulks
44
+ have no sc, and the other 1 BD and 4 HC sc samples have no bulk.
45
+
46
+ ## Method (`tools/recover_pairs.py`; v1 in `out/_v1_resid_test/`)
47
+
48
+ 1. **Identity genes, learnt from other cohorts.**
49
+ - Cohorts: Imperial ESKD, Liu 2021, Ji 2025 and Giroux 2022, using identity-pass pairs from
50
+ `paired_finetune_intradataset_20260927/data/*/pairs.csv`.
51
+ - Each donor's bulk at one timepoint is paired with its sc at a **different** timepoint, so only stable donor traits
52
+ can correlate, not the state of one blood draw.
53
+ - Per gene: Pearson across donors between bulk and sc, after centring within modality and group. Cohorts are
54
+ combined by Fisher z. chrY and sex-linked genes are excluded.
55
+ - The top genes are known strong genetic regulators of expression: UTS2, S100B, TMEM176A/B, ERAP2, DDX43, MDGA1,
56
+ FAM118A, FOLR3, GSTM3/4, HLA-DQB1 (`out/identity_genes_all4.csv`).
57
+ 2. **Score.** On those genes: log2(CPM + 1), centred within modality and disease group, then Pearson bulk × sc.
58
+ 3. **Decision rule.** Mutual best plus a statistic ≥ threshold. The threshold allows ≤ 1 % wrong calls per sc column
59
+ over 1,000 Behçet-shaped draws: bulk 9 case + 10 control, sc 4 + 4, 3 case donors in both.
60
+ - Primary test: Ye 2026 sepsis, which was never used to learn genes and is, like Behçet, a Chinese cohort with one
61
+ 10x library per donor.
62
+ - Replication: Liu 2021, scored with genes learnt without Liu.
63
+ - Primary rule, chosen as the best sepsis sensitivity: G = 50, r_best ≥ 0.658.
64
+
65
+ | control (Behçet-shaped draws) | true pair ranked first | sensitivity at ≤ 1 % wrong calls (best rule) |
66
+ |---|---:|---:|
67
+ | sepsis | 79–85 % | 55 % (G50 r_best) |
68
+ | Liu | 95–99 % | 89 % (G200 r_best) |
69
+
70
+ Sepsis is the harder control. A few outlier samples correlate with everyone: bulk Abd-PS20, and sc Abd-S102, whose own
71
+ pair correlates at r −0.10. The double-centred statistic was added to handle this; it calls all three Behçet pairs.
72
+
73
+ **What was done after seeing Behçet scores (disclosed).**
74
+ - v1 (residual top-variance genes, 20 draws) printed the same three best matches. It was dropped because its sepsis
75
+ power was low (true pair first in 60–70 % of draws).
76
+ - The first v2 test (50 draws, margin_z only) also printed Behçet scores. The r_best and double-centred statistics were
77
+ added after that, and the primary rule was then chosen on the controls only.
78
+ - The three best matches never changed across v1, v2 or any rule.
79
+
80
+ ## Files
81
+
82
+ - `tools/recover_pairs.py`: identity genes, calibration, Behçet scores and calls.
83
+ - `tools/unblind_ages.py`: age comparison, rule agreement, gene view and `out/pairs_recovered.csv`.
84
+ - `out/`: `calibration_summary.csv`, `RULE.json`, `behcet_calls.csv` (every rule), `behcet_scores_G*.csv`,
85
+ `sex_check.csv`, `unblind_ages.csv`, `pairs_recovered.csv`, `UNBLIND.json`, `SUMMARY.json`.
86
+ - `logs/`.
87
+
88
+ The datalake pairing files are not modified. The pairs live here as an expression-identity result.
89
+
90
+ ## What this allows for LoRA
91
+
92
+ - **Training data:** LoRA can be fine-tuned on the 3 BD pairs; there is no HC pair. Our smallest validated setting is
93
+ 5 pairs per cohort.
94
+ - **Test data:** cells can be generated from the 6 unpaired BD bulks and the 10 HC bulks, and their C1Q-high monocyte
95
+ share compared.
96
+ - **Limitations:**
97
+ - Only 3 training pairs.
98
+ - The model sees only BD cells during fine-tuning.
99
+ - The ground truth is still the 4 vs 4 sc donors, who are mostly not the bulk donors, plus the authors' flow
100
+ cytometry (38 vs 38, other people).
paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/logs/unblind_ages.log ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ sc_sample sc_age sc_sex_geo sc_sex_chrY most_frequent_best_bulk rules_with_this_best rules rules_mutual_with_this_best rules_calling_it primary_rule_called primary_r_best primary_threshold best_bulk_age best_bulk_sex_geo age_equal bulk_samples_with_same_age
2
+ PBMC_BD_1 30 male male HC3 6 9 3 0 False 0.334801 0.658138 18 male False BDU2
3
+ PBMC_BD_2 31 male male BDV4 9 9 9 8 True 0.784209 0.658138 31 male True BDV4
4
+ PBMC_BD_3 58 male male BDV5 9 9 9 6 False 0.555423 0.658138 58 male True BDV5;HC5
5
+ PBMC_BD_4 41 male male BDV6 9 9 9 8 True 0.745590 0.658138 41 male True BDV6;HC7
6
+ PBMC_HC_1 30 female female HC6 9 9 9 3 False 0.510968 0.658138 18 male False BDU2
7
+ PBMC_HC_2 31 male female HC2 9 9 9 0 False 0.379576 0.658138 47 male False BDV4
8
+ PBMC_HC_3 58 male male BDU2 3 9 3 0 False 0.249928 0.658138 30 male False BDV5;HC5
9
+ PBMC_HC_4 41 male male HC8 9 9 9 0 False 0.417010 0.658138 36 male False BDV6;HC7
10
+ sc_sample sc_gsm bulk_sample bulk_gsm age rules_calling_it rules primary_rule_called primary_r_best tier
11
+ PBMC_BD_2 GSM5953768 BDV4 GSM5951805 31 8 9 True 0.784209 expression_identity_called_by_primary_rule+age_equal
12
+ PBMC_BD_3 GSM5953769 BDV5 GSM5951806 58 6 9 False 0.555423 expression_best_and_mutual_in_all_rules+age_equal
13
+ PBMC_BD_4 GSM5953770 BDV6 GSM5951807 41 8 9 True 0.745590 expression_identity_called_by_primary_rule+age_equal
14
+ P(all same-age by chance) = 0.0005831753899985419
15
+ genes stat called
16
+ G100 dc_margin PBMC_BD_2->BDV4;PBMC_BD_3->BDV5;PBMC_BD_4->BDV6
17
+ G100 margin_z PBMC_BD_2->BDV4;PBMC_BD_3->BDV5;PBMC_BD_4->BDV6;PBMC_HC_1->HC6
18
+ G100 r_best PBMC_BD_2->BDV4;PBMC_BD_4->BDV6
19
+ G200 dc_margin PBMC_BD_2->BDV4;PBMC_BD_3->BDV5;PBMC_BD_4->BDV6
20
+ G200 margin_z PBMC_BD_2->BDV4;PBMC_BD_3->BDV5;PBMC_BD_4->BDV6;PBMC_HC_1->HC6
21
+ G200 r_best PBMC_BD_2->BDV4
22
+ G50 dc_margin PBMC_BD_2->BDV4;PBMC_BD_3->BDV5;PBMC_BD_4->BDV6
23
+ G50 margin_z PBMC_BD_3->BDV5;PBMC_BD_4->BDV6;PBMC_HC_1->HC6
24
+ G50 r_best PBMC_BD_2->BDV4;PBMC_BD_4->BDV6
25
+ sc_sample
26
+ PBMC_BD_2 S100B, FOLR3, ZNF683, XRRA1, PF4V1, HLA-DQB2
27
+ PBMC_BD_3 TMEM176A, PRSS21, XRRA1, MARCO, AGAP1
28
+ PBMC_BD_4 MDGA1, ZNF683, LGALS2, GLB1L
29
+ PBMC_HC_1 FOLR3
30
+ primary rule: G50 r_best 0.6581375985244046
paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/logs/v2_draws1000.log ADDED
@@ -0,0 +1,124 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'imperial_eskd': 158, 'liu2021': 61, 'ji2025_chili': 33, 'giroux2022': 24, 'ye2026_sepsis': 82}
2
+ rank gene_symbol fisher_z r_imperial_eskd r_liu2021 r_ji2025_chili r_giroux2022
3
+ 1 UTS2 1.396 0.902 0.867 0.885 0.864
4
+ 2 S100B 1.284 0.881 0.888 0.814 0.791
5
+ 3 TMEM176B 1.273 0.452 0.932 0.932 0.957
6
+ 4 TMEM176A 1.236 0.460 0.899 0.882 0.982
7
+ 5 ERAP2 1.198 0.845 0.733 0.787 0.930
8
+ 6 DDX43 1.189 0.830 0.888 0.696 0.861
9
+ 7 MDGA1 1.147 0.834 0.853 0.753 0.790
10
+ 8 FAM118A 1.125 0.651 0.706 0.932 0.911
11
+ 9 FOLR3 1.083 0.777 0.828 0.780 0.799
12
+ 10 SLC4A10 0.994 0.656 0.733 0.745 0.918
13
+ 11 MYOM2 0.977 0.882 0.563 0.057 0.941
14
+ 12 FADS2 0.976 0.820 0.690 0.746 0.649
15
+ 13 ZNF683 0.954 0.854 0.608 0.779 0.455
16
+ 14 GPAT2 0.945 0.846 0.238 0.757 0.843
17
+ 15 SMDT1 0.934 0.569 0.956 0.302 0.763
18
+ 16 POMC 0.924 0.825 0.649 0.639 0.668
19
+ 17 CFH 0.912 0.818 0.792 0.081 0.843
20
+ 18 GSTM4 0.889 0.709 0.779 0.652 0.675
21
+ 19 GSTM3 0.873 0.677 0.542 0.778 0.815
22
+ 20 LGALS2 0.867 0.652 0.755 0.771 0.603
23
+ 21 ACCS 0.864 0.762 0.861 0.536 0.285
24
+ 22 PRSS21 0.848 0.790 0.516 0.656 0.681
25
+ 23 GLB1L 0.845 0.449 0.816 0.814 0.687
26
+ 24 XRRA1 0.830 0.575 0.824 0.761 0.477
27
+ 25 RGPD2 0.829 0.421 0.849 0.808 0.598
28
+ 26 HLA-DQB1 0.819 0.482 0.858 0.781 0.467
29
+ 27 COL5A3 0.814 0.780 0.654 0.518 0.576
30
+ 28 LTK 0.806 0.234 0.668 0.683 0.965
31
+ 29 SAMD12 0.795 0.647 0.742 0.529 0.710
32
+ 30 NT5C3B 0.790 0.760 0.650 0.393 0.687
33
+ cohort genes stat threshold sensitivity wrong_call_rate draws_with_any_wrong_call top1_true_rate_shared mutual_true_rate_shared all_3_called_rate draws
34
+ liu2021 G100 dc_margin 0.2610 0.8107 0.0100 0.079 0.9893 0.9790 0.539 1000
35
+ liu2021 G100 margin_z 1.4574 0.5830 0.0100 0.079 0.9897 0.9793 0.179 1000
36
+ liu2021 G100 r_best 0.5168 0.8707 0.0100 0.080 0.9897 0.9793 0.666 1000
37
+ liu2021 G200 dc_margin 0.2511 0.8533 0.0100 0.078 0.9917 0.9860 0.621 1000
38
+ liu2021 G200 margin_z 1.4060 0.6770 0.0100 0.077 0.9917 0.9863 0.302 1000
39
+ liu2021 G200 r_best 0.4812 0.8903 0.0100 0.078 0.9917 0.9863 0.711 1000
40
+ liu2021 G50 dc_margin 0.2977 0.6210 0.0100 0.080 0.9517 0.9320 0.238 1000
41
+ liu2021 G50 margin_z 1.3619 0.4000 0.0100 0.079 0.9507 0.9307 0.058 1000
42
+ liu2021 G50 r_best 0.6252 0.7327 0.0099 0.079 0.9507 0.9307 0.412 1000
43
+ ye2026_sepsis G100 dc_margin 0.2620 0.4613 0.0100 0.079 0.8497 0.8383 0.127 1000
44
+ ye2026_sepsis G100 margin_z 1.2503 0.2947 0.0100 0.078 0.8450 0.8347 0.031 1000
45
+ ye2026_sepsis G100 r_best 0.6748 0.4640 0.0100 0.079 0.8450 0.8347 0.124 1000
46
+ ye2026_sepsis G200 dc_margin 0.2544 0.3990 0.0100 0.077 0.7983 0.7750 0.093 1000
47
+ ye2026_sepsis G200 margin_z 1.1005 0.2837 0.0100 0.077 0.7913 0.7697 0.026 1000
48
+ ye2026_sepsis G200 r_best 0.6989 0.3333 0.0100 0.079 0.7913 0.7697 0.066 1000
49
+ ye2026_sepsis G50 dc_margin 0.2884 0.4440 0.0100 0.077 0.8417 0.8287 0.111 1000
50
+ ye2026_sepsis G50 margin_z 1.2755 0.2977 0.0100 0.075 0.8360 0.8230 0.036 1000
51
+ ye2026_sepsis G50 r_best 0.6581 0.5507 0.0100 0.078 0.8360 0.8230 0.192 1000
52
+ genes stat primary sc_sample best_bulk r_best r_second mutual value threshold called sc_chrY_cpm
53
+ G100 margin_z False PBMC_BD_1 BDV3 0.189 0.167 False 0.164 1.250 False 1072.253
54
+ G100 margin_z False PBMC_BD_2 BDV4 0.770 0.299 True 1.699 1.250 True 1010.525
55
+ G100 margin_z False PBMC_BD_3 BDV5 0.606 0.198 True 1.975 1.250 True 885.544
56
+ G100 margin_z False PBMC_BD_4 BDV6 0.702 0.231 True 1.973 1.250 True 943.439
57
+ G100 margin_z False PBMC_HC_1 HC6 0.400 0.162 True 1.644 1.250 True 0.351
58
+ G100 margin_z False PBMC_HC_2 HC2 0.333 0.295 True 0.218 1.250 False 0.803
59
+ G100 margin_z False PBMC_HC_3 BDU2 0.314 0.209 True 0.658 1.250 False 969.488
60
+ G100 margin_z False PBMC_HC_4 HC8 0.302 0.206 True 0.531 1.250 False 997.048
61
+ G100 r_best False PBMC_BD_1 BDV3 0.189 0.167 False 0.189 0.675 False 1072.253
62
+ G100 r_best False PBMC_BD_2 BDV4 0.770 0.299 True 0.770 0.675 True 1010.525
63
+ G100 r_best False PBMC_BD_3 BDV5 0.606 0.198 True 0.606 0.675 False 885.544
64
+ G100 r_best False PBMC_BD_4 BDV6 0.702 0.231 True 0.702 0.675 True 943.439
65
+ G100 r_best False PBMC_HC_1 HC6 0.400 0.162 True 0.400 0.675 False 0.351
66
+ G100 r_best False PBMC_HC_2 HC2 0.333 0.295 True 0.333 0.675 False 0.803
67
+ G100 r_best False PBMC_HC_3 BDU2 0.314 0.209 True 0.314 0.675 False 969.488
68
+ G100 r_best False PBMC_HC_4 HC8 0.302 0.206 True 0.302 0.675 False 997.048
69
+ G100 dc_margin False PBMC_BD_1 BDV3 0.189 0.167 False 0.025 0.262 False 1072.253
70
+ G100 dc_margin False PBMC_BD_2 BDV4 0.770 0.299 True 0.474 0.262 True 1010.525
71
+ G100 dc_margin False PBMC_BD_3 BDV5 0.606 0.198 True 0.403 0.262 True 885.544
72
+ G100 dc_margin False PBMC_BD_4 BDV6 0.702 0.231 True 0.469 0.262 True 943.439
73
+ G100 dc_margin False PBMC_HC_1 HC6 0.400 0.162 True 0.225 0.262 False 0.351
74
+ G100 dc_margin False PBMC_HC_2 HC2 0.333 0.295 True 0.033 0.262 False 0.803
75
+ G100 dc_margin False PBMC_HC_3 BDU2 0.314 0.209 True 0.097 0.262 False 969.488
76
+ G100 dc_margin False PBMC_HC_4 HC8 0.302 0.206 True 0.100 0.262 False 997.048
77
+ G50 margin_z False PBMC_BD_1 HC3 0.335 0.185 False 0.909 1.275 False 1072.253
78
+ G50 margin_z False PBMC_BD_2 BDV4 0.784 0.376 True 1.200 1.275 False 1010.525
79
+ G50 margin_z False PBMC_BD_3 BDV5 0.555 0.258 True 1.432 1.275 True 885.544
80
+ G50 margin_z False PBMC_BD_4 BDV6 0.746 0.356 True 1.381 1.275 True 943.439
81
+ G50 margin_z False PBMC_HC_1 HC6 0.511 0.226 True 1.462 1.275 True 0.351
82
+ G50 margin_z False PBMC_HC_2 HC2 0.380 0.361 True 0.081 1.275 False 0.803
83
+ G50 margin_z False PBMC_HC_3 BDV1 0.250 0.249 True 0.004 1.275 False 969.488
84
+ G50 margin_z False PBMC_HC_4 HC8 0.417 0.264 True 0.613 1.275 False 997.048
85
+ G50 r_best True PBMC_BD_1 HC3 0.335 0.185 False 0.335 0.658 False 1072.253
86
+ G50 r_best True PBMC_BD_2 BDV4 0.784 0.376 True 0.784 0.658 True 1010.525
87
+ G50 r_best True PBMC_BD_3 BDV5 0.555 0.258 True 0.555 0.658 False 885.544
88
+ G50 r_best True PBMC_BD_4 BDV6 0.746 0.356 True 0.746 0.658 True 943.439
89
+ G50 r_best True PBMC_HC_1 HC6 0.511 0.226 True 0.511 0.658 False 0.351
90
+ G50 r_best True PBMC_HC_2 HC2 0.380 0.361 True 0.380 0.658 False 0.803
91
+ G50 r_best True PBMC_HC_3 BDV1 0.250 0.249 True 0.250 0.658 False 969.488
92
+ G50 r_best True PBMC_HC_4 HC8 0.417 0.264 True 0.417 0.658 False 997.048
93
+ G50 dc_margin False PBMC_BD_1 HC3 0.335 0.185 False 0.122 0.288 False 1072.253
94
+ G50 dc_margin False PBMC_BD_2 BDV4 0.784 0.376 True 0.415 0.288 True 1010.525
95
+ G50 dc_margin False PBMC_BD_3 BDV5 0.555 0.258 True 0.296 0.288 True 885.544
96
+ G50 dc_margin False PBMC_BD_4 BDV6 0.746 0.356 True 0.398 0.288 True 943.439
97
+ G50 dc_margin False PBMC_HC_1 HC6 0.511 0.226 True 0.263 0.288 False 0.351
98
+ G50 dc_margin False PBMC_HC_2 HC2 0.380 0.361 True 0.023 0.288 False 0.803
99
+ G50 dc_margin False PBMC_HC_3 BDV1 0.250 0.249 True 0.006 0.288 False 969.488
100
+ G50 dc_margin False PBMC_HC_4 HC8 0.417 0.264 True 0.151 0.288 False 997.048
101
+ G200 margin_z False PBMC_BD_1 HC3 0.211 0.196 True 0.110 1.100 False 1072.253
102
+ G200 margin_z False PBMC_BD_2 BDV4 0.729 0.281 True 1.799 1.100 True 1010.525
103
+ G200 margin_z False PBMC_BD_3 BDV5 0.633 0.269 True 1.583 1.100 True 885.544
104
+ G200 margin_z False PBMC_BD_4 BDV6 0.680 0.137 True 2.625 1.100 True 943.439
105
+ G200 margin_z False PBMC_HC_1 HC6 0.311 0.139 True 1.424 1.100 True 0.351
106
+ G200 margin_z False PBMC_HC_2 HC2 0.303 0.254 True 0.347 1.100 False 0.803
107
+ G200 margin_z False PBMC_HC_3 HC4 0.226 0.187 True 0.285 1.100 False 969.488
108
+ G200 margin_z False PBMC_HC_4 HC8 0.233 0.202 True 0.219 1.100 False 997.048
109
+ G200 r_best False PBMC_BD_1 HC3 0.211 0.196 True 0.211 0.699 False 1072.253
110
+ G200 r_best False PBMC_BD_2 BDV4 0.729 0.281 True 0.729 0.699 True 1010.525
111
+ G200 r_best False PBMC_BD_3 BDV5 0.633 0.269 True 0.633 0.699 False 885.544
112
+ G200 r_best False PBMC_BD_4 BDV6 0.680 0.137 True 0.680 0.699 False 943.439
113
+ G200 r_best False PBMC_HC_1 HC6 0.311 0.139 True 0.311 0.699 False 0.351
114
+ G200 r_best False PBMC_HC_2 HC2 0.303 0.254 True 0.303 0.699 False 0.803
115
+ G200 r_best False PBMC_HC_3 HC4 0.226 0.187 True 0.226 0.699 False 969.488
116
+ G200 r_best False PBMC_HC_4 HC8 0.233 0.202 True 0.233 0.699 False 997.048
117
+ G200 dc_margin False PBMC_BD_1 HC3 0.211 0.196 True 0.009 0.254 False 1072.253
118
+ G200 dc_margin False PBMC_BD_2 BDV4 0.729 0.281 True 0.451 0.254 True 1010.525
119
+ G200 dc_margin False PBMC_BD_3 BDV5 0.633 0.269 True 0.364 0.254 True 885.544
120
+ G200 dc_margin False PBMC_BD_4 BDV6 0.680 0.137 True 0.534 0.254 True 943.439
121
+ G200 dc_margin False PBMC_HC_1 HC6 0.311 0.139 True 0.166 0.254 False 0.351
122
+ G200 dc_margin False PBMC_HC_2 HC2 0.303 0.254 True 0.046 0.254 False 0.803
123
+ G200 dc_margin False PBMC_HC_3 HC4 0.226 0.187 True 0.045 0.254 False 969.488
124
+ G200 dc_margin False PBMC_HC_4 HC8 0.233 0.202 True 0.028 0.254 False 997.048
paper-releases/pbisc-diffusion/updates/20261004T003810Z-behcet-pair-recovery/diffusion/dataset/server_1/model8/plans/behcet_pair_recovery_20261004/logs/v2_test_draws50.log ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {'imperial_eskd': 158, 'liu2021': 61, 'ji2025_chili': 33, 'giroux2022': 24, 'ye2026_sepsis': 82}
2
+ rank gene_symbol fisher_z r_imperial_eskd r_liu2021 r_ji2025_chili r_giroux2022
3
+ 1 UTS2 1.396 0.902 0.867 0.885 0.864
4
+ 2 S100B 1.284 0.881 0.888 0.814 0.791
5
+ 3 TMEM176B 1.273 0.452 0.932 0.932 0.957
6
+ 4 TMEM176A 1.236 0.460 0.899 0.882 0.982
7
+ 5 ERAP2 1.198 0.845 0.733 0.787 0.930
8
+ 6 DDX43 1.189 0.830 0.888 0.696 0.861
9
+ 7 MDGA1 1.147 0.834 0.853 0.753 0.790
10
+ 8 FAM118A 1.125 0.651 0.706 0.932 0.911
11
+ 9 FOLR3 1.083 0.777 0.828 0.780 0.799
12
+ 10 SLC4A10 0.994 0.656 0.733 0.745 0.918
13
+ 11 MYOM2 0.977 0.882 0.563 0.057 0.941
14
+ 12 FADS2 0.976 0.820 0.690 0.746 0.649
15
+ 13 ZNF683 0.954 0.854 0.608 0.779 0.455
16
+ 14 GPAT2 0.945 0.846 0.238 0.757 0.843
17
+ 15 SMDT1 0.934 0.569 0.956 0.302 0.763
18
+ 16 POMC 0.924 0.825 0.649 0.639 0.668
19
+ 17 CFH 0.912 0.818 0.792 0.081 0.843
20
+ 18 GSTM4 0.889 0.709 0.779 0.652 0.675
21
+ 19 GSTM3 0.873 0.677 0.542 0.778 0.815
22
+ 20 LGALS2 0.867 0.652 0.755 0.771 0.603
23
+ 21 ACCS 0.864 0.762 0.861 0.536 0.285
24
+ 22 PRSS21 0.848 0.790 0.516 0.656 0.681
25
+ 23 GLB1L 0.845 0.449 0.816 0.814 0.687
26
+ 24 XRRA1 0.830 0.575 0.824 0.761 0.477
27
+ 25 RGPD2 0.829 0.421 0.849 0.808 0.598
28
+ 26 HLA-DQB1 0.819 0.482 0.858 0.781 0.467
29
+ 27 COL5A3 0.814 0.780 0.654 0.518 0.576
30
+ 28 LTK 0.806 0.234 0.668 0.683 0.965
31
+ 29 SAMD12 0.795 0.647 0.742 0.529 0.710
32
+ 30 NT5C3B 0.790 0.760 0.650 0.393 0.687
33
+ cohort genes threshold sensitivity wrong_call_rate draws_with_any_wrong_call top1_true_rate_shared mutual_true_rate_shared all_3_called_rate draws
34
+ liu2021 G100 1.2685 0.7533 0.01 0.08 1.0000 1.0000 0.46 50
35
+ liu2021 G200 1.3312 0.7933 0.01 0.08 1.0000 1.0000 0.52 50
36
+ liu2021 G50 1.3382 0.3733 0.01 0.08 0.9667 0.9533 0.04 50
37
+ ye2026_sepsis G100 1.2974 0.2067 0.01 0.08 0.8267 0.8200 0.02 50
38
+ ye2026_sepsis G200 1.0265 0.3000 0.01 0.08 0.7667 0.7200 0.00 50
39
+ ye2026_sepsis G50 1.2629 0.2733 0.01 0.08 0.8333 0.8133 0.08 50
40
+ genes sc_sample best_bulk second_bulk r_best r_second mutual margin threshold called
41
+ G100 PBMC_BD_1 BDV3 BDU1 0.189 0.167 False 0.164 1.297 False
42
+ G100 PBMC_BD_2 BDV4 BDV3 0.770 0.299 True 1.699 1.297 True
43
+ G100 PBMC_BD_3 BDV5 HC2 0.606 0.198 True 1.975 1.297 True
44
+ G100 PBMC_BD_4 BDV6 HC3 0.702 0.231 True 1.973 1.297 True
45
+ G100 PBMC_HC_1 HC6 HC10 0.400 0.162 True 1.644 1.297 True
46
+ G100 PBMC_HC_2 HC2 BDU1 0.333 0.295 True 0.218 1.297 False
47
+ G100 PBMC_HC_3 BDU2 HC1 0.314 0.209 True 0.658 1.297 False
48
+ G100 PBMC_HC_4 HC8 HC5 0.302 0.206 True 0.531 1.297 False
49
+ G50 PBMC_BD_1 HC3 BDV3 0.335 0.185 False 0.909 1.263 False
50
+ G50 PBMC_BD_2 BDV4 BDV3 0.784 0.376 True 1.200 1.263 False
51
+ G50 PBMC_BD_3 BDV5 HC9 0.555 0.258 True 1.432 1.263 True
52
+ G50 PBMC_BD_4 BDV6 HC3 0.746 0.356 True 1.381 1.263 True
53
+ G50 PBMC_HC_1 HC6 HC10 0.511 0.226 True 1.462 1.263 True
54
+ G50 PBMC_HC_2 HC2 BDU1 0.380 0.361 True 0.081 1.263 False
55
+ G50 PBMC_HC_3 BDV1 BDU2 0.250 0.249 True 0.004 1.263 False
56
+ G50 PBMC_HC_4 HC8 HC5 0.417 0.264 True 0.613 1.263 False
57
+ G200 PBMC_BD_1 HC3 BDV3 0.211 0.196 True 0.110 1.026 False
58
+ G200 PBMC_BD_2 BDV4 BDV3 0.729 0.281 True 1.799 1.026 True
59
+ G200 PBMC_BD_3 BDV5 HC2 0.633 0.269 True 1.583 1.026 True
60
+ G200 PBMC_BD_4 BDV6 HC7 0.680 0.137 True 2.625 1.026 True
61
+ G200 PBMC_HC_1 HC6 BDU2 0.311 0.139 True 1.424 1.026 True
62
+ G200 PBMC_HC_2 HC2 BDU1 0.303 0.254 True 0.347 1.026 False
63
+ G200 PBMC_HC_3 HC4 BDU2 0.226 0.187 True 0.285 1.026 False
64
+ G200 PBMC_HC_4 HC8 BDV4 0.233 0.202 True 0.219 1.026 False